[R] How to clear R memory in a for loop

[Prof Brian Ripley]

On 20/10/2014 17:53, John McKown wrote:
> On Mon, Oct 20, 2014 at 10:30 AM, Dimitri Liakhovitski <
> dimitri.liakhovitski at gmail.com> wrote:
>
>> Dear Rers,
>>
>> I am trying to run a for-loop in R.
>> During each iteration I read in an mp3 file and do some basic processing.
>> If I do what I need to do for each file one by one - it works fine.
>> But once I start running a loop, it soon runs out of memory and says: can't
>> allocate a vector of size...
>> In each iteration of my loop I always overwrite the previously created
>> object and do gc().
>>
>> Any hints on how to fight this?
>>
>> Thanks a lot!
>>
>>
>>
> ​Please don't use HTML for messages.
>
> What occurs to me, from reading the other replies, is that perhaps within
> the loop you are causing other objects to be allocated. And that can be
> done just by doing a simple assignment, so it may not be obvious. What this
> can do is cause what we called a "sand bar" in the old days. That's where
> you allocate a big chunk of memory for an object. Say this take up 1/2 of
> your available space. You now create a small object. This object is
> _probably_ right next to the large object. You now release the large
> object. Your apparent free space is now almost what it was at the
> beginning. But when you try to allocate another large object which is, say,
> 2/3 of the maximum space, you can't because that small object is sitting
> right in the middle of our memory space. So you _can_ allocate 2 large
> objects which are 1/3 your free space size, but not 1 object which is 2/3
> of the free space size. Which can lead to your type of situation.
>
> This is just a SWAG based on some experience in other systems. Most
> "garbage collection" do _not_ do memory consolidation. I don't know about
> R.​
>
>
That is true of R (except for the early days which did have a moving 
garbage collector).

However 'your available space' is not the amount of RAM you have but the 
process address space.  The latter is enormous on any 64-bit OS, so 
'memory fragmentation' (as this is termed) is a thing of the past except 
for those limited to many-years-old OSes.


-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Emeritus Professor of Applied Statistics, University of Oxford
1 South Parks Road, Oxford OX1 3TG, UK